Introduction of Celex Family Sensor and Event/Frame/Optical-flow Hybrid Processing

Event: CVPR 2019 Workshop on Event-based Vision · Duration: 24 min · ▶ Watch on YouTube

Abstract

This presentation introduces the Celex family of event-based sensors, highlighting their advantages over traditional frame-based sensors, such as low latency and reduced data redundancy. The speaker details the history of Celex sensor development, from early prototypes to automotive-qualified megapixel sensors. Key innovations include a novel event packet format (X,Y,A,T) that provides both in-pixel time-stamps and logarithmic grey-level values, enabling robust motion and appearance modeling. The presentation also covers on-chip optical flow capabilities and a hybrid processing approach that seamlessly integrates full-picture, event, and optical flow modes for enhanced performance in various applications, particularly in the automotive sector.

Speakers

  • Shoushun Chen — CelePixel Technology Co. Ltd

Talks (1)

  • 00:00:00 — Shoushun Chen: Introduction of Celex Family Sensor and Event/Frame/Optical-flow Hybrid Processing
    • A presentation on Celex event-based sensors, their unique features like in-pixel time-stamping and hybrid processing capabilities, and their application in areas like automotive.

Key Takeaways

  • Celex event-based sensors offer significantly lower latency and reduced redundant data compared to traditional frame-based sensors, boosting the response time of vision systems.
  • The unique (X,Y,A,T) event packet format provides both accurate in-pixel time-stamps and pixel logarithmic grey-level values, allowing for comprehensive spatial-temporal information for appearance and motion models.
  • Celex sensors integrate on-chip optical flow and support multi-mode auto-cycling between full-picture, event, and optical flow outputs, enhancing processing efficiency and adaptability.
  • The in-pixel time-stamp mechanism ensures accurate, jitter-free event timing, crucial for high-speed motion tracking and precise event correlation.
  • Hybrid processing, combining different sensor output modes, is being actively developed to address complex computational vision challenges, particularly in demanding automotive applications like lane detection, vehicle tracking, and human tracking.

Methods / Models / Datasets Mentioned

  • Celex Family Sensor
  • Celex-V
  • Celex-X
  • DAVIS
  • ATIS
  • VI-SLAM

Topics

Event-based sensors · Hybrid processing · Optical flow · In-pixel time-stamp · Low latency · Dynamic range · Automotive applications · Computational vision

Notes

Open for commentary — connections to other work, critiques, follow-up reading.